AWCR: Adaptive and Weighted Collaborative Representations for Face Super-Resolution with Context Residual-Learning Conference Paper uri icon

abstract

  • © Springer International Publishing AG, part of Springer Nature 2018. Owing to the ill-posed nature of the image super-resolution (SR) problem, learning-based approaches typically employ regularization terms in the representation. Current local-patch based face SR approaches weight representation coefficients to obtain adaptive and accurate priors. However, they ignore the fact that heteroskedasticity generally exists both in the observed data and representation coefficients. In this paper, we present a novel adaptive and weighted representation framework for face SR to further exploit adaptive and accurate prior information for different content inputs. Moreover, we enrich patch priors by sampling context patches, and learn the residual high-frequency components for better reconstruction performance. Experiments on the CAS-PEAL-R1 face database show that our proposed approach outperforms state-of-the-arts that include other deep learning based methods.

author list (cited authors)

  • Lu, T., Pan, L., Wang, J., Zhang, Y., Wang, Z., & Xiong, Z.

citation count

  • 1

complete list of authors

  • Lu, Tao||Pan, Lanlan||Wang, Jiaming||Zhang, Yanduo||Wang, Zhongyuan||Xiong, Zixiang

publication date

  • May 2018